[arin-ppml] 2010-8: Rework of IPv6 assignment criteria

[William Herrin]

On Sat, Sep 18, 2010 at 8:00 PM, David Farmer <farmer at umn.edu> wrote:
>> Presumably, the reason for the difference is not that end user
>> organizations are not capable of understanding the HD-ratio concept. And
>> anyway, if it is difficult to understand, ARIN can be asked to produce
>> explanatory materials.
>
> I want to refer people to RFC 3194: "The Host-Density Ratio for Address
> Assignment Efficiency: An update on the H ratio" and section 6.7 of the
> NRPM.
>
> https://www.arin.net/policy/nrpm.html#six7
>
> Actually, while I believe many people do not understand HD-Ratio, especially
> end-users, that isn't the only issue. if it were, then "explanatory
> materials" could maybe solve the problem.

RFC 3194 basically says this:

We threw some formulas around and when we tried "log(addresses in
use)/log(total address space) " it seemed to produce some consistent
results for "too dense" across more than one historical addressing
scheme. We found that:

log(addresses in use)/log(total address space) < 0.80 works reasonably well and
log(addresses in use)/log(total address space) > 0.87 doesn't work at all.

In terms of addressing policy, it implies: get the next address
allocation when your HD Ratio approaches 0.80.

The basis for these numbers does not appear to be especially well
sourced. Only three addressing systems were evaluated: two telephone
numbering schemes and DECnet.

It is, however, a testable prediction, the stuff any science is based
on. RFC 3194 predicts that we can successfully employ no more than
214M IPv4 unicast addresses on the IPv4 Internet, or about 5.7% of the
address space.

Finding numbers on actual IPv4 address utilization proved difficult. I
did find some numbers from CAIDA. They weren't quite on point, but
they suggested that current utilization stands around 11%. If correct
then the HD Ratio is disproven as a metric for assessing address
utilization as 11% is far greater than the predicted maximum 5.7% and
we're still functioning without excessive pain.

As I said, those numbers weren't entirely on point so I can't use them
to say they absolutely disprove RFC 3194. It does, however, make
3194's claims more dubious.


The ARIN NRPM then takes this relatively straightforward theory and
does something incomprehensible with it. It talks about "efficiency
level is between 20% to 50% for most ISPs and LIRs when based on a
0.94 HD Ratio." And then it talks about /56 counts. But the theory
behind HD Ratio is based on addresses deployed, not subnetworks. And
it predicts a much lower HD ratio. And percentages of some HD ratio
don't even make sense in context; you're supposed to use the ratio
directly; the calculation replaces percentages.


So, our current IPv6 addressing policy starts with an unproven and
dubious theory that was then altered in arbitrary ways that totally
removed it from any scientific basis.

Such is our current basis for what's supposed to be reasonable IPv6
addressing policy.

Regards,
Bill Herrin



-- 
William D. Herrin ................ herrin at dirtside.com  bill at herrin.us
3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004